agent.py

import os

from dotenv import load_dotenv
from langchain_community.vectorstores import Chroma
from langchain_core.messages import HumanMessage, SystemMessage, ToolMessage
from langchain_huggingface import (ChatHuggingFace, HuggingFaceEmbeddings,
                                   HuggingFaceEndpoint)
from langgraph.graph import START, MessagesState, StateGraph
from langgraph.prebuilt import ToolNode, tools_condition

from tools import (absolute, add, analyze_excel_file, arvix_search,
                   audio_transcription, compound_interest, convert_temperature,
                   divide, exponential, factorial, floor_divide,
                   get_current_time_in_timezone, greatest_common_divisor,
                   is_prime, least_common_multiple, logarithm, modulus,
                   multiply, percentage_calculator, power, python_code_parser,
                   roman_calculator_converter, square_root, subtract,
                   web_search, wiki_search)

# Load Constants
load_dotenv()
HUGGINGFACEHUB_API_TOKEN = os.getenv("HUGGINGFACEHUB_API_TOKEN")


tools = [
    multiply, add, subtract, power, divide, modulus,
    square_root, floor_divide, absolute, logarithm,
    exponential, web_search, roman_calculator_converter,
    get_current_time_in_timezone, compound_interest,
    convert_temperature, factorial, greatest_common_divisor,
    is_prime, least_common_multiple, percentage_calculator,
    wiki_search, analyze_excel_file, arvix_search, audio_transcription, python_code_parser
]

# Load system prompt
system_prompt = """
You are a helpful assistant tasked with answering questions using a set of tools. 
Now, I will ask you a question. Report your thoughts, and finish your answer with the following template: 
FINAL ANSWER: [YOUR FINAL ANSWER]. 
YOUR FINAL ANSWER should be a number OR as few words as possible OR a comma separated list of numbers and/or strings. If you are asked for a number, don't use comma to write your number neither use units such as $ or percent sign unless specified otherwise. If you are asked for a string, don't use articles, neither abbreviations (e.g. for cities), and write the digits in plain text unless specified otherwise. If you are asked for a comma separated list, apply the above rules depending of whether the element to be put in the list is a number or a string.
Your answer should only start with "FINAL ANSWER: ", then follows with the answer. 
"""

# System message
sys_msg = SystemMessage(content=system_prompt)


def get_vector_store(persist_directory="chroma_db"):
    """
    Initializes and returns a Chroma vector store.
    If the database exists, it loads it. If not, it creates it,
    adds some initial documents, and persists them.
    """
    embedding_function = HuggingFaceEmbeddings(model_name="sentence-transformers/all-mpnet-base-v2")

    if os.path.exists(persist_directory) and os.listdir(persist_directory):
        print("Loading existing vector store...")
        vector_store = Chroma(
            persist_directory=persist_directory,
            embedding_function=embedding_function
        )
    else:
        print("Creating new vector store...")
        os.makedirs(persist_directory, exist_ok=True)
        # Example documents to add
        initial_documents = [
            "The Principle of Double Effect is an ethical theory that distinguishes between the intended and foreseen consequences of an action.",
            "St. Thomas Aquinas is often associated with the development of the Principle of Double Effect.",
            "LangGraph is a library for building stateful, multi-actor applications with LLMs.",
            "Chroma is a vector database used for storing and retrieving embeddings."
        ]
        vector_store = Chroma.from_texts(
            texts=initial_documents,
            embedding=embedding_function,
            persist_directory=persist_directory
        )
        # No need to call persist() when using from_texts with a persist_directory

    return vector_store

# --- Initialize Vector Store and Retriever ---
vector_store = get_vector_store()
retriever_component = vector_store.as_retriever(
    search_type="mmr",  # Use Maximum Marginal Relevance for diverse results
    search_kwargs={'k': 2, 'lambda_mult': 0.5}  # Retrieve 2 documents
)


def build_graph():
    """Build the graph"""

    # First create the HuggingFaceEndpoint
    llm_endpoint = HuggingFaceEndpoint(
        repo_id="gemini/gemini-2.0-flash",
        #"Qwen/Qwen2.5-Coder-32B-Instruct",
        #huggingfacehub_api_token=HUGGINGFACEHUB_API_TOKEN,
        api_key=os.getenv("GEMINI_API_KEY"),
        temperature=0.3,
        max_new_tokens=512,
        timeout=60,
    )

    # Then wrap it with ChatHuggingFace to get chat model functionality
    llm = ChatHuggingFace(llm=llm_endpoint)

    # Bind tools to LLM
    llm_with_tools = llm.bind_tools(tools)

    # --- Nodes ---
    def assistant(state: MessagesState):
        """Assistant node"""
        # Prepend the system message to the state
        messages_with_system_prompt = [sys_msg] + state["messages"]
        return {"messages": [llm_with_tools.invoke(messages_with_system_prompt)]}

    def retriever_node(state: MessagesState):
        """
        Retrieves relevant documents from the vector store based on the latest human message.
        """
        last_human_message = state["messages"][-1].content
        retrieved_docs = retriever_component.invoke(last_human_message)
        
        if retrieved_docs:
            retrieved_context = "\n\n".join([doc.page_content for doc in retrieved_docs])
            # Create a ToolMessage to hold the retrieved context
            context_message = ToolMessage(
                content=f"Retrieved context from vector store:\n\n{retrieved_context}",
                tool_call_id="retriever" # A descriptive ID
            )
            return {"messages": [context_message]}
        
        return {"messages": []}

    # --- Graph Definition ---
    builder = StateGraph(MessagesState)
    builder.add_node("retriever", retriever_node)
    builder.add_node("assistant", assistant)
    builder.add_node("tools", ToolNode(tools))

    builder.add_edge(START, "retriever")
    builder.add_edge("retriever", "assistant")
    builder.add_conditional_edges("assistant", tools_condition)
    builder.add_edge("tools", "assistant")

    # Compile graph
    return builder.compile()


# test
if __name__ == "__main__":
    question = "When was a picture of St. Thomas Aquinas first added to the Wikipedia page on the Principle of double effect?"
    # Build the graph
    graph = build_graph()
    # Run the graph
    messages = [HumanMessage(content=question)]
    # The initial state for the graph
    initial_state = {"messages": messages}
    
    # Invoke the graph stream to see the steps
    for s in graph.stream(initial_state, stream_mode="values"):
        message = s["messages"][-1]
        if isinstance(message, ToolMessage):
            print("---RETRIEVED CONTEXT---")
            print(message.content)
            print("-----------------------")
        else:
            message.pretty_print()